blob: 889287019599a861d90a51d6808cfb2d032fc255 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/block_dev.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/init.h>
9#include <linux/mm.h>
10#include <linux/fcntl.h>
11#include <linux/slab.h>
12#include <linux/kmod.h>
13#include <linux/major.h>
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -070014#include <linux/device_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/highmem.h>
16#include <linux/blkdev.h>
17#include <linux/module.h>
18#include <linux/blkpg.h>
19#include <linux/buffer_head.h>
Nick Piggin585d3bc2009-02-25 10:44:19 +010020#include <linux/pagevec.h>
David Howells811d7362006-08-29 19:06:09 +010021#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/mpage.h>
23#include <linux/mount.h>
24#include <linux/uio.h>
25#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070026#include <linux/log2.h>
Catalin Marinas2e1483c2009-06-11 13:24:13 +010027#include <linux/kmemleak.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020029#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31struct bdev_inode {
32 struct block_device bdev;
33 struct inode vfs_inode;
34};
35
Adrian Bunk4c54ac62008-02-18 13:48:31 +010036static const struct address_space_operations def_blk_aops;
37
Linus Torvalds1da177e2005-04-16 15:20:36 -070038static inline struct bdev_inode *BDEV_I(struct inode *inode)
39{
40 return container_of(inode, struct bdev_inode, vfs_inode);
41}
42
43inline struct block_device *I_BDEV(struct inode *inode)
44{
45 return &BDEV_I(inode)->bdev;
46}
47
48EXPORT_SYMBOL(I_BDEV);
49
Dave Chinnera5491e02010-10-21 11:49:26 +110050/*
51 * move the inode from it's current bdi to the a new bdi. if the inode is dirty
52 * we need to move it onto the dirty list of @dst so that the inode is always
53 * on the right list.
54 */
55static void bdev_inode_switch_bdi(struct inode *inode,
56 struct backing_dev_info *dst)
57{
58 spin_lock(&inode_lock);
59 inode->i_data.backing_dev_info = dst;
60 if (inode->i_state & I_DIRTY)
Nick Piggin7ccf19a2010-10-21 11:49:30 +110061 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
Dave Chinnera5491e02010-10-21 11:49:26 +110062 spin_unlock(&inode_lock);
63}
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065static sector_t max_block(struct block_device *bdev)
66{
67 sector_t retval = ~((sector_t)0);
68 loff_t sz = i_size_read(bdev->bd_inode);
69
70 if (sz) {
71 unsigned int size = block_size(bdev);
72 unsigned int sizebits = blksize_bits(size);
73 retval = (sz >> sizebits);
74 }
75 return retval;
76}
77
Peter Zijlstraf9a14392007-05-06 14:49:55 -070078/* Kill _all_ buffers and pagecache , dirty or not.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -070079static void kill_bdev(struct block_device *bdev)
80{
Peter Zijlstraf9a14392007-05-06 14:49:55 -070081 if (bdev->bd_inode->i_mapping->nrpages == 0)
82 return;
83 invalidate_bh_lrus();
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
85}
86
87int set_blocksize(struct block_device *bdev, int size)
88{
89 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070090 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 return -EINVAL;
92
93 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -040094 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 return -EINVAL;
96
97 /* Don't change the size if it is same as current */
98 if (bdev->bd_block_size != size) {
99 sync_blockdev(bdev);
100 bdev->bd_block_size = size;
101 bdev->bd_inode->i_blkbits = blksize_bits(size);
102 kill_bdev(bdev);
103 }
104 return 0;
105}
106
107EXPORT_SYMBOL(set_blocksize);
108
109int sb_set_blocksize(struct super_block *sb, int size)
110{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 if (set_blocksize(sb->s_bdev, size))
112 return 0;
113 /* If we get here, we know size is power of two
114 * and it's value is between 512 and PAGE_SIZE */
115 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800116 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 return sb->s_blocksize;
118}
119
120EXPORT_SYMBOL(sb_set_blocksize);
121
122int sb_min_blocksize(struct super_block *sb, int size)
123{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400124 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 if (size < minsize)
126 size = minsize;
127 return sb_set_blocksize(sb, size);
128}
129
130EXPORT_SYMBOL(sb_min_blocksize);
131
132static int
133blkdev_get_block(struct inode *inode, sector_t iblock,
134 struct buffer_head *bh, int create)
135{
136 if (iblock >= max_block(I_BDEV(inode))) {
137 if (create)
138 return -EIO;
139
140 /*
141 * for reads, we're just trying to fill a partial page.
142 * return a hole, they will have to call get_block again
143 * before they can fill it, and they will get -EIO at that
144 * time
145 */
146 return 0;
147 }
148 bh->b_bdev = I_BDEV(inode);
149 bh->b_blocknr = iblock;
150 set_buffer_mapped(bh);
151 return 0;
152}
153
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800154static int
155blkdev_get_blocks(struct inode *inode, sector_t iblock,
156 struct buffer_head *bh, int create)
157{
158 sector_t end_block = max_block(I_BDEV(inode));
159 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
160
161 if ((iblock + max_blocks) > end_block) {
162 max_blocks = end_block - iblock;
163 if ((long)max_blocks <= 0) {
164 if (create)
165 return -EIO; /* write fully beyond EOF */
166 /*
167 * It is a read which is fully beyond EOF. We return
168 * a !buffer_mapped buffer
169 */
170 max_blocks = 0;
171 }
172 }
173
174 bh->b_bdev = I_BDEV(inode);
175 bh->b_blocknr = iblock;
176 bh->b_size = max_blocks << inode->i_blkbits;
177 if (max_blocks)
178 set_buffer_mapped(bh);
179 return 0;
180}
181
182static ssize_t
183blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
184 loff_t offset, unsigned long nr_segs)
185{
186 struct file *file = iocb->ki_filp;
187 struct inode *inode = file->f_mapping->host;
188
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +0200189 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
190 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800191}
192
Jan Kara5cee5812009-04-27 16:43:51 +0200193int __sync_blockdev(struct block_device *bdev, int wait)
194{
195 if (!bdev)
196 return 0;
197 if (!wait)
198 return filemap_flush(bdev->bd_inode->i_mapping);
199 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
200}
201
Nick Piggin585d3bc2009-02-25 10:44:19 +0100202/*
203 * Write out and wait upon all the dirty data associated with a block
204 * device via its mapping. Does not take the superblock lock.
205 */
206int sync_blockdev(struct block_device *bdev)
207{
Jan Kara5cee5812009-04-27 16:43:51 +0200208 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100209}
210EXPORT_SYMBOL(sync_blockdev);
211
212/*
213 * Write out and wait upon all dirty data associated with this
214 * device. Filesystem data as well as the underlying block
215 * device. Takes the superblock lock.
216 */
217int fsync_bdev(struct block_device *bdev)
218{
219 struct super_block *sb = get_super(bdev);
220 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200221 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100222 drop_super(sb);
223 return res;
224 }
225 return sync_blockdev(bdev);
226}
Al Viro47e44912009-04-01 07:07:16 -0400227EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100228
229/**
230 * freeze_bdev -- lock a filesystem and force it into a consistent state
231 * @bdev: blockdevice to lock
232 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100233 * If a superblock is found on this device, we take the s_umount semaphore
234 * on it to make sure nobody unmounts until the snapshot creation is done.
235 * The reference counter (bd_fsfreeze_count) guarantees that only the last
236 * unfreeze process can unfreeze the frozen filesystem actually when multiple
237 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
238 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
239 * actually.
240 */
241struct super_block *freeze_bdev(struct block_device *bdev)
242{
243 struct super_block *sb;
244 int error = 0;
245
246 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200247 if (++bdev->bd_fsfreeze_count > 1) {
248 /*
249 * We don't even need to grab a reference - the first call
250 * to freeze_bdev grab an active reference and only the last
251 * thaw_bdev drops it.
252 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100253 sb = get_super(bdev);
Christoph Hellwig45042302009-08-03 23:28:35 +0200254 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100255 mutex_unlock(&bdev->bd_fsfreeze_mutex);
256 return sb;
257 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100258
Christoph Hellwig45042302009-08-03 23:28:35 +0200259 sb = get_active_super(bdev);
260 if (!sb)
261 goto out;
Josef Bacik18e9e512010-03-23 10:34:56 -0400262 error = freeze_super(sb);
263 if (error) {
264 deactivate_super(sb);
265 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200266 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400267 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100268 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400269 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200270 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100271 sync_blockdev(bdev);
272 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200273 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100274}
275EXPORT_SYMBOL(freeze_bdev);
276
277/**
278 * thaw_bdev -- unlock filesystem
279 * @bdev: blockdevice to unlock
280 * @sb: associated superblock
281 *
282 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
283 */
284int thaw_bdev(struct block_device *bdev, struct super_block *sb)
285{
Christoph Hellwig45042302009-08-03 23:28:35 +0200286 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100287
288 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200289 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400290 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100291
Christoph Hellwig45042302009-08-03 23:28:35 +0200292 error = 0;
293 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400294 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100295
Christoph Hellwig45042302009-08-03 23:28:35 +0200296 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400297 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200298
Josef Bacik18e9e512010-03-23 10:34:56 -0400299 error = thaw_super(sb);
300 if (error) {
301 bdev->bd_fsfreeze_count++;
302 mutex_unlock(&bdev->bd_fsfreeze_mutex);
303 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100304 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400305out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100306 mutex_unlock(&bdev->bd_fsfreeze_mutex);
307 return 0;
308}
309EXPORT_SYMBOL(thaw_bdev);
310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
312{
313 return block_write_full_page(page, blkdev_get_block, wbc);
314}
315
316static int blkdev_readpage(struct file * file, struct page * page)
317{
318 return block_read_full_page(page, blkdev_get_block);
319}
320
Nick Piggin6272b5a2007-10-16 01:25:04 -0700321static int blkdev_write_begin(struct file *file, struct address_space *mapping,
322 loff_t pos, unsigned len, unsigned flags,
323 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200325 return block_write_begin(mapping, pos, len, flags, pagep,
326 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327}
328
Nick Piggin6272b5a2007-10-16 01:25:04 -0700329static int blkdev_write_end(struct file *file, struct address_space *mapping,
330 loff_t pos, unsigned len, unsigned copied,
331 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700333 int ret;
334 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
335
336 unlock_page(page);
337 page_cache_release(page);
338
339 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340}
341
342/*
343 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800344 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 * so we compute the size by hand (just as in block_read/write above)
346 */
347static loff_t block_llseek(struct file *file, loff_t offset, int origin)
348{
349 struct inode *bd_inode = file->f_mapping->host;
350 loff_t size;
351 loff_t retval;
352
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800353 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 size = i_size_read(bd_inode);
355
356 switch (origin) {
357 case 2:
358 offset += size;
359 break;
360 case 1:
361 offset += file->f_pos;
362 }
363 retval = -EINVAL;
364 if (offset >= 0 && offset <= size) {
365 if (offset != file->f_pos) {
366 file->f_pos = offset;
367 }
368 retval = offset;
369 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800370 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 return retval;
372}
373
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200374int blkdev_fsync(struct file *filp, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400376 struct inode *bd_inode = filp->f_mapping->host;
377 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100378 int error;
379
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400380 /*
381 * There is no need to serialise calls to blkdev_issue_flush with
382 * i_mutex and doing so causes performance issues with concurrent
383 * O_SYNC writers to a block device.
384 */
385 mutex_unlock(&bd_inode->i_mutex);
386
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200387 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100388 if (error == -EOPNOTSUPP)
389 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400390
391 mutex_lock(&bd_inode->i_mutex);
392
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100393 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700395EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
397/*
398 * pseudo-fs
399 */
400
401static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800402static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
404static struct inode *bdev_alloc_inode(struct super_block *sb)
405{
Christoph Lametere94b1762006-12-06 20:33:17 -0800406 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 if (!ei)
408 return NULL;
409 return &ei->vfs_inode;
410}
411
Nick Pigginfa0d7e32011-01-07 17:49:49 +1100412static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
Nick Pigginfa0d7e32011-01-07 17:49:49 +1100414 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 struct bdev_inode *bdi = BDEV_I(inode);
416
Nick Pigginfa0d7e32011-01-07 17:49:49 +1100417 INIT_LIST_HEAD(&inode->i_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 kmem_cache_free(bdev_cachep, bdi);
419}
420
Nick Pigginfa0d7e32011-01-07 17:49:49 +1100421static void bdev_destroy_inode(struct inode *inode)
422{
423 call_rcu(&inode->i_rcu, bdev_i_callback);
424}
425
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700426static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427{
428 struct bdev_inode *ei = (struct bdev_inode *) foo;
429 struct block_device *bdev = &ei->bdev;
430
Christoph Lametera35afb82007-05-16 22:10:57 -0700431 memset(bdev, 0, sizeof(*bdev));
432 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700433 INIT_LIST_HEAD(&bdev->bd_inodes);
434 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100435#ifdef CONFIG_SYSFS
436 INIT_LIST_HEAD(&bdev->bd_holder_disks);
437#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700438 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800439 /* Initialize mutex for freeze. */
440 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441}
442
443static inline void __bd_forget(struct inode *inode)
444{
445 list_del_init(&inode->i_devices);
446 inode->i_bdev = NULL;
447 inode->i_mapping = &inode->i_data;
448}
449
Al Virob57922d2010-06-07 14:34:48 -0400450static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451{
452 struct block_device *bdev = &BDEV_I(inode)->bdev;
453 struct list_head *p;
Al Virob57922d2010-06-07 14:34:48 -0400454 truncate_inode_pages(&inode->i_data, 0);
455 invalidate_inode_buffers(inode); /* is it needed here? */
456 end_writeback(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 spin_lock(&bdev_lock);
458 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
459 __bd_forget(list_entry(p, struct inode, i_devices));
460 }
461 list_del_init(&bdev->bd_list);
462 spin_unlock(&bdev_lock);
463}
464
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800465static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 .statfs = simple_statfs,
467 .alloc_inode = bdev_alloc_inode,
468 .destroy_inode = bdev_destroy_inode,
469 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400470 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471};
472
Al Viro51139ad2010-07-25 23:47:46 +0400473static struct dentry *bd_mount(struct file_system_type *fs_type,
474 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475{
Al Viroc74a1cb2011-01-12 16:59:34 -0500476 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477}
478
479static struct file_system_type bd_type = {
480 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400481 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 .kill_sb = kill_anon_super,
483};
484
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800485struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
487void __init bdev_cache_init(void)
488{
489 int err;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800490 struct vfsmount *bd_mnt;
491
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800493 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
494 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900495 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 err = register_filesystem(&bd_type);
497 if (err)
498 panic("Cannot register bdev pseudo-fs");
499 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 if (IS_ERR(bd_mnt))
501 panic("Cannot create bdev pseudo-fs");
Catalin Marinas2e1483c2009-06-11 13:24:13 +0100502 /*
503 * This vfsmount structure is only used to obtain the
504 * blockdev_superblock, so tell kmemleak not to report it.
505 */
506 kmemleak_not_leak(bd_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
508}
509
510/*
511 * Most likely _very_ bad one - but then it's hardly critical for small
512 * /dev and can be fixed when somebody will need really large one.
513 * Keep in mind that it will be fed through icache hash function too.
514 */
515static inline unsigned long hash(dev_t dev)
516{
517 return MAJOR(dev)+MINOR(dev);
518}
519
520static int bdev_test(struct inode *inode, void *data)
521{
522 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
523}
524
525static int bdev_set(struct inode *inode, void *data)
526{
527 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
528 return 0;
529}
530
531static LIST_HEAD(all_bdevs);
532
533struct block_device *bdget(dev_t dev)
534{
535 struct block_device *bdev;
536 struct inode *inode;
537
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800538 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 bdev_test, bdev_set, &dev);
540
541 if (!inode)
542 return NULL;
543
544 bdev = &BDEV_I(inode)->bdev;
545
546 if (inode->i_state & I_NEW) {
547 bdev->bd_contains = NULL;
548 bdev->bd_inode = inode;
549 bdev->bd_block_size = (1 << inode->i_blkbits);
550 bdev->bd_part_count = 0;
551 bdev->bd_invalidated = 0;
552 inode->i_mode = S_IFBLK;
553 inode->i_rdev = dev;
554 inode->i_bdev = bdev;
555 inode->i_data.a_ops = &def_blk_aops;
556 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
557 inode->i_data.backing_dev_info = &default_backing_dev_info;
558 spin_lock(&bdev_lock);
559 list_add(&bdev->bd_list, &all_bdevs);
560 spin_unlock(&bdev_lock);
561 unlock_new_inode(inode);
562 }
563 return bdev;
564}
565
566EXPORT_SYMBOL(bdget);
567
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200568/**
569 * bdgrab -- Grab a reference to an already referenced block device
570 * @bdev: Block device to grab a reference to.
571 */
572struct block_device *bdgrab(struct block_device *bdev)
573{
Al Viro7de9c6e2010-10-23 11:11:40 -0400574 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200575 return bdev;
576}
577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578long nr_blockdev_pages(void)
579{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700580 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 long ret = 0;
582 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700583 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 ret += bdev->bd_inode->i_mapping->nrpages;
585 }
586 spin_unlock(&bdev_lock);
587 return ret;
588}
589
590void bdput(struct block_device *bdev)
591{
592 iput(bdev->bd_inode);
593}
594
595EXPORT_SYMBOL(bdput);
596
597static struct block_device *bd_acquire(struct inode *inode)
598{
599 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 spin_lock(&bdev_lock);
602 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700603 if (bdev) {
Al Viro7de9c6e2010-10-23 11:11:40 -0400604 ihold(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 spin_unlock(&bdev_lock);
606 return bdev;
607 }
608 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 bdev = bdget(inode->i_rdev);
611 if (bdev) {
612 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700613 if (!inode->i_bdev) {
614 /*
Al Viro7de9c6e2010-10-23 11:11:40 -0400615 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700616 * and it's released in clear_inode() of inode.
617 * So, we can access it via ->i_mapping always
618 * without igrab().
619 */
Al Viro7de9c6e2010-10-23 11:11:40 -0400620 ihold(bdev->bd_inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700621 inode->i_bdev = bdev;
622 inode->i_mapping = bdev->bd_inode->i_mapping;
623 list_add(&inode->i_devices, &bdev->bd_inodes);
624 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 spin_unlock(&bdev_lock);
626 }
627 return bdev;
628}
629
630/* Call when you free inode */
631
632void bd_forget(struct inode *inode)
633{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700634 struct block_device *bdev = NULL;
635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700637 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800638 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700639 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700643
644 if (bdev)
645 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900648/**
649 * bd_may_claim - test whether a block device can be claimed
650 * @bdev: block device of interest
651 * @whole: whole block device containing @bdev, may equal @bdev
652 * @holder: holder trying to claim @bdev
653 *
654 * Test whther @bdev can be claimed by @holder.
655 *
656 * CONTEXT:
657 * spin_lock(&bdev_lock).
658 *
659 * RETURNS:
660 * %true if @bdev can be claimed, %false otherwise.
661 */
662static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
663 void *holder)
664{
665 if (bdev->bd_holder == holder)
666 return true; /* already a holder */
667 else if (bdev->bd_holder != NULL)
668 return false; /* held by someone else */
669 else if (bdev->bd_contains == bdev)
670 return true; /* is a whole device which isn't held */
671
Tejun Heoe525fd82010-11-13 11:55:17 +0100672 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900673 return true; /* is a partition of a device that is being partitioned */
674 else if (whole->bd_holder != NULL)
675 return false; /* is a partition of a held device */
676 else
677 return true; /* is a partition of an un-held device */
678}
679
680/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900681 * bd_prepare_to_claim - prepare to claim a block device
682 * @bdev: block device of interest
683 * @whole: the whole device containing @bdev, may equal @bdev
684 * @holder: holder trying to claim @bdev
685 *
686 * Prepare to claim @bdev. This function fails if @bdev is already
687 * claimed by another holder and waits if another claiming is in
688 * progress. This function doesn't actually claim. On successful
689 * return, the caller has ownership of bd_claiming and bd_holder[s].
690 *
691 * CONTEXT:
692 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
693 * it multiple times.
694 *
695 * RETURNS:
696 * 0 if @bdev can be claimed, -EBUSY otherwise.
697 */
698static int bd_prepare_to_claim(struct block_device *bdev,
699 struct block_device *whole, void *holder)
700{
701retry:
702 /* if someone else claimed, fail */
703 if (!bd_may_claim(bdev, whole, holder))
704 return -EBUSY;
705
Tejun Heoe75aa852010-08-04 17:59:39 +0200706 /* if claiming is already in progress, wait for it to finish */
707 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900708 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
709 DEFINE_WAIT(wait);
710
711 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
712 spin_unlock(&bdev_lock);
713 schedule();
714 finish_wait(wq, &wait);
715 spin_lock(&bdev_lock);
716 goto retry;
717 }
718
719 /* yay, all mine */
720 return 0;
721}
722
723/**
724 * bd_start_claiming - start claiming a block device
725 * @bdev: block device of interest
726 * @holder: holder trying to claim @bdev
727 *
728 * @bdev is about to be opened exclusively. Check @bdev can be opened
729 * exclusively and mark that an exclusive open is in progress. Each
730 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000731 * either bd_finish_claiming() or bd_abort_claiming() (which do not
732 * fail).
733 *
734 * This function is used to gain exclusive access to the block device
735 * without actually causing other exclusive open attempts to fail. It
736 * should be used when the open sequence itself requires exclusive
737 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900738 *
739 * CONTEXT:
740 * Might sleep.
741 *
742 * RETURNS:
743 * Pointer to the block device containing @bdev on success, ERR_PTR()
744 * value on failure.
745 */
746static struct block_device *bd_start_claiming(struct block_device *bdev,
747 void *holder)
748{
749 struct gendisk *disk;
750 struct block_device *whole;
751 int partno, err;
752
753 might_sleep();
754
755 /*
756 * @bdev might not have been initialized properly yet, look up
757 * and grab the outer block device the hard way.
758 */
759 disk = get_gendisk(bdev->bd_dev, &partno);
760 if (!disk)
761 return ERR_PTR(-ENXIO);
762
763 whole = bdget_disk(disk, 0);
Nick Piggincf342572010-05-26 01:50:21 +1000764 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900765 put_disk(disk);
766 if (!whole)
767 return ERR_PTR(-ENOMEM);
768
769 /* prepare to claim, if successful, mark claiming in progress */
770 spin_lock(&bdev_lock);
771
772 err = bd_prepare_to_claim(bdev, whole, holder);
773 if (err == 0) {
774 whole->bd_claiming = holder;
775 spin_unlock(&bdev_lock);
776 return whole;
777 } else {
778 spin_unlock(&bdev_lock);
779 bdput(whole);
780 return ERR_PTR(err);
781 }
782}
783
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800784#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100785struct bd_holder_disk {
786 struct list_head list;
787 struct gendisk *disk;
788 int refcnt;
789};
790
791static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
792 struct gendisk *disk)
793{
794 struct bd_holder_disk *holder;
795
796 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
797 if (holder->disk == disk)
798 return holder;
799 return NULL;
800}
801
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700802static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800803{
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700804 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800805}
806
807static void del_symlink(struct kobject *from, struct kobject *to)
808{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800809 sysfs_remove_link(from, kobject_name(to));
810}
811
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800812/**
Tejun Heoe09b4572010-11-13 11:55:17 +0100813 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
814 * @bdev: the claimed slave bdev
815 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500816 *
Tejun Heo49731ba2011-01-14 18:43:57 +0100817 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
818 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100819 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500820 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100821 * - from "slaves" directory of the holder @disk to the claimed @bdev
822 * - from "holders" directory of the @bdev to the holder @disk
823 *
824 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
825 * passed to bd_link_disk_holder(), then:
826 *
827 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
828 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
829 *
830 * The caller must have claimed @bdev before calling this function and
831 * ensure that both @bdev and @disk are valid during the creation and
832 * lifetime of these symlinks.
833 *
834 * CONTEXT:
835 * Might sleep.
836 *
837 * RETURNS:
838 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500839 */
Tejun Heoe09b4572010-11-13 11:55:17 +0100840int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500841{
Tejun Heo49731ba2011-01-14 18:43:57 +0100842 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100843 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800844
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800845 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500846
Tejun Heo49731ba2011-01-14 18:43:57 +0100847 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -0700848
Tejun Heoe09b4572010-11-13 11:55:17 +0100849 /* FIXME: remove the following once add_disk() handles errors */
850 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
851 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -0700852
Tejun Heo49731ba2011-01-14 18:43:57 +0100853 holder = bd_find_holder_disk(bdev, disk);
854 if (holder) {
855 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +0100856 goto out_unlock;
857 }
858
Tejun Heo49731ba2011-01-14 18:43:57 +0100859 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
860 if (!holder) {
861 ret = -ENOMEM;
862 goto out_unlock;
863 }
864
865 INIT_LIST_HEAD(&holder->list);
866 holder->disk = disk;
867 holder->refcnt = 1;
868
869 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
870 if (ret)
871 goto out_free;
872
873 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
874 if (ret)
875 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +0100876 /*
877 * bdev could be deleted beneath us which would implicitly destroy
878 * the holder directory. Hold on to it.
879 */
880 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100881
882 list_add(&holder->list, &bdev->bd_holder_disks);
883 goto out_unlock;
884
885out_del:
886 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
887out_free:
888 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +0100889out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800890 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +0100891 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800892}
Tejun Heoe09b4572010-11-13 11:55:17 +0100893EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800894
Tejun Heo49731ba2011-01-14 18:43:57 +0100895/**
896 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
897 * @bdev: the calimed slave bdev
898 * @disk: the holding disk
899 *
900 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
901 *
902 * CONTEXT:
903 * Might sleep.
904 */
905void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800906{
Tejun Heo49731ba2011-01-14 18:43:57 +0100907 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100908
Tejun Heo49731ba2011-01-14 18:43:57 +0100909 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800910
Tejun Heo49731ba2011-01-14 18:43:57 +0100911 holder = bd_find_holder_disk(bdev, disk);
912
913 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
914 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
915 del_symlink(bdev->bd_part->holder_dir,
916 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +0100917 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100918 list_del_init(&holder->list);
919 kfree(holder);
920 }
921
922 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800923}
Tejun Heo49731ba2011-01-14 18:43:57 +0100924EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800925#endif
926
Andrew Patterson0c002c22008-09-04 14:27:20 -0600927/**
Andrew Patterson56ade442008-09-04 14:27:40 -0600928 * flush_disk - invalidates all buffer-cache entries on a disk
929 *
930 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -0800931 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -0600932 *
933 * Invalidates all buffer-cache entries on a disk. It should be called
934 * when a disk has been changed -- either by a media change or online
935 * resize.
936 */
NeilBrown93b270f2011-02-24 17:25:47 +1100937static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -0600938{
NeilBrown93b270f2011-02-24 17:25:47 +1100939 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -0600940 char name[BDEVNAME_SIZE] = "";
941
942 if (bdev->bd_disk)
943 disk_name(bdev->bd_disk, 0, name);
944 printk(KERN_WARNING "VFS: busy inodes on changed media or "
945 "resized disk %s\n", name);
946 }
947
948 if (!bdev->bd_disk)
949 return;
950 if (disk_partitionable(bdev->bd_disk))
951 bdev->bd_invalidated = 1;
952}
953
954/**
Randy Dunlap57d1b532008-10-09 10:42:38 +0200955 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600956 * @disk: struct gendisk to check
957 * @bdev: struct bdev to adjust.
958 *
959 * This routine checks to see if the bdev size does not match the disk size
960 * and adjusts it if it differs.
961 */
962void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
963{
964 loff_t disk_size, bdev_size;
965
966 disk_size = (loff_t)get_capacity(disk) << 9;
967 bdev_size = i_size_read(bdev->bd_inode);
968 if (disk_size != bdev_size) {
969 char name[BDEVNAME_SIZE];
970
971 disk_name(disk, 0, name);
972 printk(KERN_INFO
973 "%s: detected capacity change from %lld to %lld\n",
974 name, bdev_size, disk_size);
975 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +1100976 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600977 }
978}
979EXPORT_SYMBOL(check_disk_size_change);
980
981/**
Randy Dunlap57d1b532008-10-09 10:42:38 +0200982 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -0600983 * @disk: struct gendisk to be revalidated
984 *
985 * This routine is a wrapper for lower-level driver's revalidate_disk
986 * call-backs. It is used to do common pre and post operations needed
987 * for all revalidate_disk operations.
988 */
989int revalidate_disk(struct gendisk *disk)
990{
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600991 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -0600992 int ret = 0;
993
994 if (disk->fops->revalidate_disk)
995 ret = disk->fops->revalidate_disk(disk);
996
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600997 bdev = bdget_disk(disk, 0);
998 if (!bdev)
999 return ret;
1000
1001 mutex_lock(&bdev->bd_mutex);
1002 check_disk_size_change(disk, bdev);
1003 mutex_unlock(&bdev->bd_mutex);
1004 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001005 return ret;
1006}
1007EXPORT_SYMBOL(revalidate_disk);
1008
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009/*
1010 * This routine checks whether a removable media has been changed,
1011 * and invalidates all buffer-cache-entries in that case. This
1012 * is a relatively slow routine, so we have to try to minimize using
1013 * it. Thus it is called only upon a 'mount' or 'open'. This
1014 * is the best way of combining speed and utility, I think.
1015 * People changing diskettes in the middle of an operation deserve
1016 * to lose :-)
1017 */
1018int check_disk_change(struct block_device *bdev)
1019{
1020 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001021 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001022 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
Tejun Heo77ea8872010-12-08 20:57:37 +01001024 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1025 DISK_EVENT_EJECT_REQUEST);
1026 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 return 0;
1028
NeilBrown93b270f2011-02-24 17:25:47 +11001029 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 if (bdops->revalidate_disk)
1031 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 return 1;
1033}
1034
1035EXPORT_SYMBOL(check_disk_change);
1036
1037void bd_set_size(struct block_device *bdev, loff_t size)
1038{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001039 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 bdev->bd_inode->i_size = size;
1042 while (bsize < PAGE_CACHE_SIZE) {
1043 if (size & bsize)
1044 break;
1045 bsize <<= 1;
1046 }
1047 bdev->bd_block_size = bsize;
1048 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1049}
1050EXPORT_SYMBOL(bd_set_size);
1051
Al Viro9a1c3542008-02-22 20:40:24 -05001052static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001053
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001054/*
1055 * bd_mutex locking:
1056 *
1057 * mutex_lock(part->bd_mutex)
1058 * mutex_lock_nested(whole->bd_mutex, 1)
1059 */
1060
Al Viro572c4892007-10-08 13:24:05 -04001061static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001064 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001065 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001066 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
Al Viro572c4892007-10-08 13:24:05 -04001068 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001069 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001070 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001071 perm |= MAY_WRITE;
1072 /*
1073 * hooks: /n/, see "layering violations".
1074 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001075 if (!for_part) {
1076 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1077 if (ret != 0) {
1078 bdput(bdev);
1079 return ret;
1080 }
Al Viro82666022008-08-01 05:32:04 -04001081 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001082
NeilBrownd3374822009-01-09 08:31:10 +11001083 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001084
Tejun Heo89f97492008-11-05 10:21:06 +01001085 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001086 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001087 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001088 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
NeilBrown6796bf52006-12-08 02:36:16 -08001090 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 if (!bdev->bd_openers) {
1092 bdev->bd_disk = disk;
1093 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001094 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001096
1097 ret = -ENXIO;
1098 bdev->bd_part = disk_get_part(disk, partno);
1099 if (!bdev->bd_part)
1100 goto out_clear;
1101
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001103 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001104 if (ret == -ERESTARTSYS) {
1105 /* Lost a race with 'disk' being
1106 * deleted, try again.
1107 * See md.c
1108 */
1109 disk_put_part(bdev->bd_part);
1110 bdev->bd_part = NULL;
1111 module_put(disk->fops->owner);
1112 put_disk(disk);
1113 bdev->bd_disk = NULL;
1114 mutex_unlock(&bdev->bd_mutex);
1115 goto restart;
1116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001118 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 }
1120 if (!bdev->bd_openers) {
1121 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1122 bdi = blk_get_backing_dev_info(bdev);
1123 if (bdi == NULL)
1124 bdi = &default_backing_dev_info;
Dave Chinnera5491e02010-10-21 11:49:26 +11001125 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 }
1127 if (bdev->bd_invalidated)
1128 rescan_partitions(disk, bdev);
1129 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 struct block_device *whole;
1131 whole = bdget_disk(disk, 0);
1132 ret = -ENOMEM;
1133 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001134 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001135 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001136 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001138 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 bdev->bd_contains = whole;
Dave Chinnera5491e02010-10-21 11:49:26 +11001140 bdev_inode_switch_bdi(bdev->bd_inode,
1141 whole->bd_inode->i_data.backing_dev_info);
Tejun Heo89f97492008-11-05 10:21:06 +01001142 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001143 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001144 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001146 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 }
Tejun Heo89f97492008-11-05 10:21:06 +01001148 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 }
1150 } else {
Tejun Heo0762b8b2008-08-25 19:56:12 +09001151 module_put(disk->fops->owner);
Neil Brown960cc0f2009-10-26 08:59:17 +01001152 put_disk(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001153 disk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 if (bdev->bd_contains == bdev) {
1155 if (bdev->bd_disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001156 ret = bdev->bd_disk->fops->open(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001158 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 }
1160 if (bdev->bd_invalidated)
1161 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 }
1163 }
1164 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001165 if (for_part)
1166 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001167 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 return 0;
1169
Tejun Heo0762b8b2008-08-25 19:56:12 +09001170 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001171 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001173 bdev->bd_part = NULL;
Dave Chinnera5491e02010-10-21 11:49:26 +11001174 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001176 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001178 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001179 mutex_unlock(&bdev->bd_mutex);
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001180 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001181 if (disk)
1182 module_put(disk->fops->owner);
1183 put_disk(disk);
1184 bdput(bdev);
1185
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 return ret;
1187}
1188
Tejun Heod4d77622010-11-13 11:55:18 +01001189/**
1190 * blkdev_get - open a block device
1191 * @bdev: block_device to open
1192 * @mode: FMODE_* mask
1193 * @holder: exclusive holder identifier
1194 *
1195 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1196 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1197 * @holder is invalid. Exclusive opens may nest for the same @holder.
1198 *
1199 * On success, the reference count of @bdev is unchanged. On failure,
1200 * @bdev is put.
1201 *
1202 * CONTEXT:
1203 * Might sleep.
1204 *
1205 * RETURNS:
1206 * 0 on success, -errno on failure.
1207 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001208int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209{
Tejun Heoe525fd82010-11-13 11:55:17 +01001210 struct block_device *whole = NULL;
1211 int res;
1212
1213 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1214
1215 if ((mode & FMODE_EXCL) && holder) {
1216 whole = bd_start_claiming(bdev, holder);
1217 if (IS_ERR(whole)) {
1218 bdput(bdev);
1219 return PTR_ERR(whole);
1220 }
1221 }
1222
1223 res = __blkdev_get(bdev, mode, 0);
1224
1225 if (whole) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001226 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001227 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001228 spin_lock(&bdev_lock);
1229
Tejun Heo77ea8872010-12-08 20:57:37 +01001230 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001231 BUG_ON(!bd_may_claim(bdev, whole, holder));
1232 /*
1233 * Note that for a whole device bd_holders
1234 * will be incremented twice, and bd_holder
1235 * will be set to bd_may_claim before being
1236 * set to holder
1237 */
1238 whole->bd_holders++;
1239 whole->bd_holder = bd_may_claim;
1240 bdev->bd_holders++;
1241 bdev->bd_holder = holder;
1242 }
1243
1244 /* tell others that we're done */
1245 BUG_ON(whole->bd_claiming != holder);
1246 whole->bd_claiming = NULL;
1247 wake_up_bit(&whole->bd_claiming, 0);
1248
1249 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001250
1251 /*
1252 * Block event polling for write claims. Any write
1253 * holder makes the write_holder state stick until all
1254 * are released. This is good enough and tracking
1255 * individual writeable reference is too fragile given
1256 * the way @mode is used in blkdev_get/put().
1257 */
1258 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder) {
1259 bdev->bd_write_holder = true;
1260 disk_block_events(bdev->bd_disk);
1261 }
1262
1263 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001264 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001265 }
1266
1267 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001268}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269EXPORT_SYMBOL(blkdev_get);
1270
Tejun Heod4d77622010-11-13 11:55:18 +01001271/**
1272 * blkdev_get_by_path - open a block device by name
1273 * @path: path to the block device to open
1274 * @mode: FMODE_* mask
1275 * @holder: exclusive holder identifier
1276 *
1277 * Open the blockdevice described by the device file at @path. @mode
1278 * and @holder are identical to blkdev_get().
1279 *
1280 * On success, the returned block_device has reference count of one.
1281 *
1282 * CONTEXT:
1283 * Might sleep.
1284 *
1285 * RETURNS:
1286 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1287 */
1288struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1289 void *holder)
1290{
1291 struct block_device *bdev;
1292 int err;
1293
1294 bdev = lookup_bdev(path);
1295 if (IS_ERR(bdev))
1296 return bdev;
1297
1298 err = blkdev_get(bdev, mode, holder);
1299 if (err)
1300 return ERR_PTR(err);
1301
Chuck Ebberte51900f2011-02-16 18:11:53 -05001302 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1303 blkdev_put(bdev, mode);
1304 return ERR_PTR(-EACCES);
1305 }
1306
Tejun Heod4d77622010-11-13 11:55:18 +01001307 return bdev;
1308}
1309EXPORT_SYMBOL(blkdev_get_by_path);
1310
1311/**
1312 * blkdev_get_by_dev - open a block device by device number
1313 * @dev: device number of block device to open
1314 * @mode: FMODE_* mask
1315 * @holder: exclusive holder identifier
1316 *
1317 * Open the blockdevice described by device number @dev. @mode and
1318 * @holder are identical to blkdev_get().
1319 *
1320 * Use it ONLY if you really do not have anything better - i.e. when
1321 * you are behind a truly sucky interface and all you are given is a
1322 * device number. _Never_ to be used for internal purposes. If you
1323 * ever need it - reconsider your API.
1324 *
1325 * On success, the returned block_device has reference count of one.
1326 *
1327 * CONTEXT:
1328 * Might sleep.
1329 *
1330 * RETURNS:
1331 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1332 */
1333struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1334{
1335 struct block_device *bdev;
1336 int err;
1337
1338 bdev = bdget(dev);
1339 if (!bdev)
1340 return ERR_PTR(-ENOMEM);
1341
1342 err = blkdev_get(bdev, mode, holder);
1343 if (err)
1344 return ERR_PTR(err);
1345
1346 return bdev;
1347}
1348EXPORT_SYMBOL(blkdev_get_by_dev);
1349
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350static int blkdev_open(struct inode * inode, struct file * filp)
1351{
1352 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
1354 /*
1355 * Preserve backwards compatibility and allow large file access
1356 * even if userspace doesn't ask for it explicitly. Some mkfs
1357 * binary needs it. We might want to drop this workaround
1358 * during an unstable branch.
1359 */
1360 filp->f_flags |= O_LARGEFILE;
1361
Al Viro572c4892007-10-08 13:24:05 -04001362 if (filp->f_flags & O_NDELAY)
1363 filp->f_mode |= FMODE_NDELAY;
1364 if (filp->f_flags & O_EXCL)
1365 filp->f_mode |= FMODE_EXCL;
1366 if ((filp->f_flags & O_ACCMODE) == 3)
1367 filp->f_mode |= FMODE_WRITE_IOCTL;
1368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001370 if (bdev == NULL)
1371 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
Al Viro572c4892007-10-08 13:24:05 -04001373 filp->f_mapping = bdev->bd_inode->i_mapping;
1374
Tejun Heoe525fd82010-11-13 11:55:17 +01001375 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376}
1377
Al Viro9a1c3542008-02-22 20:40:24 -05001378static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001379{
1380 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001381 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001382 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001383
NeilBrown6796bf52006-12-08 02:36:16 -08001384 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001385 if (for_part)
1386 bdev->bd_part_count--;
1387
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001388 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001389 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001390 sync_blockdev(bdev);
1391 kill_bdev(bdev);
1392 }
1393 if (bdev->bd_contains == bdev) {
1394 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001395 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001396 }
1397 if (!bdev->bd_openers) {
1398 struct module *owner = disk->fops->owner;
1399
1400 put_disk(disk);
1401 module_put(owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001402 disk_put_part(bdev->bd_part);
1403 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001404 bdev->bd_disk = NULL;
Dave Chinnera5491e02010-10-21 11:49:26 +11001405 bdev_inode_switch_bdi(bdev->bd_inode,
1406 &default_backing_dev_info);
NeilBrown37be4122006-12-08 02:36:16 -08001407 if (bdev != bdev->bd_contains)
1408 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001409 bdev->bd_contains = NULL;
1410 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001411 mutex_unlock(&bdev->bd_mutex);
1412 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001413 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001414 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001415 return ret;
1416}
1417
Al Viro9a1c3542008-02-22 20:40:24 -05001418int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001419{
Tejun Heoe525fd82010-11-13 11:55:17 +01001420 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001421 bool bdev_free;
1422
1423 /*
1424 * Release a claim on the device. The holder fields
1425 * are protected with bdev_lock. bd_mutex is to
1426 * synchronize disk_holder unlinking.
1427 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001428 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001429 spin_lock(&bdev_lock);
1430
1431 WARN_ON_ONCE(--bdev->bd_holders < 0);
1432 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1433
1434 /* bd_contains might point to self, check in a separate step */
1435 if ((bdev_free = !bdev->bd_holders))
1436 bdev->bd_holder = NULL;
1437 if (!bdev->bd_contains->bd_holders)
1438 bdev->bd_contains->bd_holder = NULL;
1439
1440 spin_unlock(&bdev_lock);
1441
Tejun Heo77ea8872010-12-08 20:57:37 +01001442 /*
1443 * If this was the last claim, remove holder link and
1444 * unblock evpoll if it was a write holder.
1445 */
1446 if (bdev_free) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001447 if (bdev->bd_write_holder) {
1448 disk_unblock_events(bdev->bd_disk);
1449 bdev->bd_write_holder = false;
1450 } else
1451 disk_check_events(bdev->bd_disk);
1452 }
Tejun Heo6a027ef2010-11-13 11:55:17 +01001453
Tejun Heoe525fd82010-11-13 11:55:17 +01001454 mutex_unlock(&bdev->bd_mutex);
Tejun Heo77ea8872010-12-08 20:57:37 +01001455 } else
1456 disk_check_events(bdev->bd_disk);
1457
Al Viro9a1c3542008-02-22 20:40:24 -05001458 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001459}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001460EXPORT_SYMBOL(blkdev_put);
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462static int blkdev_close(struct inode * inode, struct file * filp)
1463{
1464 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
Tejun Heoe525fd82010-11-13 11:55:17 +01001465
Al Viro9a1c3542008-02-22 20:40:24 -05001466 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467}
1468
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001469static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470{
Al Viro56b26ad2008-09-19 03:17:36 -04001471 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1472 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001473
1474 /*
1475 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1476 * to updated it before every ioctl.
1477 */
Al Viro56b26ad2008-09-19 03:17:36 -04001478 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001479 mode |= FMODE_NDELAY;
1480 else
1481 mode &= ~FMODE_NDELAY;
1482
Al Viro56b26ad2008-09-19 03:17:36 -04001483 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484}
1485
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001486/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001487 * Write data to the block device. Only intended for the block device itself
1488 * and the raw driver which basically is a fake block device.
1489 *
1490 * Does not take i_mutex for the write and thus is not for general purpose
1491 * use.
1492 */
1493ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1494 unsigned long nr_segs, loff_t pos)
1495{
1496 struct file *file = iocb->ki_filp;
1497 ssize_t ret;
1498
1499 BUG_ON(iocb->ki_pos != pos);
1500
1501 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1502 if (ret > 0 || ret == -EIOCBQUEUED) {
1503 ssize_t err;
1504
1505 err = generic_write_sync(file, pos, ret);
1506 if (err < 0 && ret > 0)
1507 ret = err;
1508 }
1509 return ret;
1510}
1511EXPORT_SYMBOL_GPL(blkdev_aio_write);
1512
1513/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001514 * Try to release a page associated with block device when the system
1515 * is under memory pressure.
1516 */
1517static int blkdev_releasepage(struct page *page, gfp_t wait)
1518{
1519 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1520
1521 if (super && super->s_op->bdev_try_to_free_page)
1522 return super->s_op->bdev_try_to_free_page(super, page, wait);
1523
1524 return try_to_free_buffers(page);
1525}
1526
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001527static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 .readpage = blkdev_readpage,
1529 .writepage = blkdev_writepage,
1530 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001531 .write_begin = blkdev_write_begin,
1532 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001534 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 .direct_IO = blkdev_direct_IO,
1536};
1537
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001538const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 .open = blkdev_open,
1540 .release = blkdev_close,
1541 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001542 .read = do_sync_read,
1543 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 .aio_read = generic_file_aio_read,
Christoph Hellwigeef99382009-08-20 17:43:41 +02001545 .aio_write = blkdev_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001547 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001548 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549#ifdef CONFIG_COMPAT
1550 .compat_ioctl = compat_blkdev_ioctl,
1551#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001552 .splice_read = generic_file_splice_read,
1553 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554};
1555
1556int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1557{
1558 int res;
1559 mm_segment_t old_fs = get_fs();
1560 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001561 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 set_fs(old_fs);
1563 return res;
1564}
1565
1566EXPORT_SYMBOL(ioctl_by_bdev);
1567
1568/**
1569 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001570 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001572 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 * namespace if possible and return it. Return ERR_PTR(error)
1574 * otherwise.
1575 */
Al Viro421748e2008-08-02 01:04:36 -04001576struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577{
1578 struct block_device *bdev;
1579 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001580 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 int error;
1582
Al Viro421748e2008-08-02 01:04:36 -04001583 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 return ERR_PTR(-EINVAL);
1585
Al Viro421748e2008-08-02 01:04:36 -04001586 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 if (error)
1588 return ERR_PTR(error);
1589
Al Viro421748e2008-08-02 01:04:36 -04001590 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 error = -ENOTBLK;
1592 if (!S_ISBLK(inode->i_mode))
1593 goto fail;
1594 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001595 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 goto fail;
1597 error = -ENOMEM;
1598 bdev = bd_acquire(inode);
1599 if (!bdev)
1600 goto fail;
1601out:
Al Viro421748e2008-08-02 01:04:36 -04001602 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 return bdev;
1604fail:
1605 bdev = ERR_PTR(error);
1606 goto out;
1607}
Al Virod5686b42008-08-01 05:00:11 -04001608EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
NeilBrown93b270f2011-02-24 17:25:47 +11001610int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001611{
1612 struct super_block *sb = get_super(bdev);
1613 int res = 0;
1614
1615 if (sb) {
1616 /*
1617 * no need to lock the super, get_super holds the
1618 * read mutex so the filesystem cannot go away
1619 * under us (->put_super runs with the write lock
1620 * hold).
1621 */
1622 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001623 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001624 drop_super(sb);
1625 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001626 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001627 return res;
1628}
1629EXPORT_SYMBOL(__invalidate_device);